Skip to main content

Settings

Color Mode

Theme Skin

Background

Appearance preferences are saved in this browser only.

Environment

Current Environment Production

Built with JEKYLL_ENV=production. Changes require deployment.

Quick Links

Theme & Build

Jekyll v3.10.0
Last BuildSep 08, 07:22
Repository
bamr87/bamr87

Page Location

Page Info

Layout default
Collection dash
Path _dash/harness.md
URL /harness/
Date 2026-09-08

Harness

The six-layer harness scorecard and trip wires, plus the diagrams that illustrate how the hub's loops run across the fleet.

Table of Contents

🧬 Harness

Agent = Model + Harness. The model reasons; everything else — guides, sensors, bounded loops, memory, permissions, observability — is this repo. The scorecard and trip wires below are computed offline from the committed fleet signals by dash-gen harness (refreshed daily by fleet-pulse.yml) into _data/harness_health.yml. Thresholds live in _data/fleet.ymlharness:. Operator guide: docs/HARNESS.md.

3 / 6

Trip wires tripped

89.7%

Completion rate (floor 80)

48.3%

Effectiveness (floor 70)

28

Standing failures

Snapshot 2026-09-08 06:51 UTC · inputs: actions_usage (0.0d) fleet_triage (0.0d) issue_pipeline (0.9d) token_rotation (1.2d).

Trip wires

A tripped wire is an attention item, not an action — the doctor, issue pipeline, and rotation loops own the fixes. Every wire is listed, armed or tripped, so a quiet panel is distinguishable from a lost one.

WireState
armed stale-data all fleet signals fresh
armed pass-rate-floor fleet workflow success rate 89.7% vs floor 75%
TRIPPED waste-ceiling wasted minutes 51.7% of total vs ceiling 30%
TRIPPED cost-spike workflows averaging ≥3.0× the fleet median (0.83 min)
  • lifehacker.dev · content-factory — 53.98 min avg over 4 runs
  • law-ai · refresh-lsat-content — 37.79 min avg over 9 runs
  • it-journey · ♾️ Quest Perfection Loop — 20.03 min avg over 8 runs
  • law-ai · Integration Tests — 15.95 min avg over 25 runs
  • zer0-mistakes · Comprehensive CI Pipeline — 13.15 min avg over 41 runs
armed standing-failures 28 standing red workflows vs max 30
TRIPPED credential-overdue a fleet credential is past its rotation policy plus grace
  • ·

Scorecard

MetricValueBetterThresholdStatus
completion_rate_pct 89.7 ↑ higher 80 ok
effectiveness_pct 48.3 ↑ higher 70 warn
waste_hours 53.7 ↓ lower
cost_min_per_verified_run 3.11 ↓ lower
standing_failures 28 ↓ lower
repos_red 22 ↓ lower
escalations_open 77 ↓ lower
agent_prs_open 15 → steady
oldest_credential_age_days 85 ↓ lower

How it runs — illustrated

Interactive diagrams authored as typed JSON in diagrams/ and compiled by the vendored archify skill (tools/render-diagrams.sh). Each opens standalone with search, focus, route tracing, dark/light theme, and export. Guided chapters are in the ▶ views menu.

The six layers, mapped onto the hubOpen standalone ↗
fleet-pulse.yml — the daily loopOpen standalone ↗
Fleet signals — from GitHub to the alarm panelOpen standalone ↗
issue-pipeline.yml — labels as stateOpen standalone ↗
repo-evolution.yml — a draft PR into a submodule's own upstreamOpen standalone ↗